If you surveyed accounting partners, SMSF administrators, and SMB owners today, almost none would put "MCP support" on their software shortlist. In twelve months, that's going to look like a mistake.
Here's why.
Since November 2024, an open standard called the Model Context Protocol has gone from an Anthropic side project to the default way AI agents talk to software. The numbers are worth pausing on. MCP SDK downloads went from around 100,000 in the first month to 97 million per month by March 2026. OpenAI, Google, Microsoft, and Salesforce all shipped support inside thirteen months. Gartner expects 40% of enterprise applications to include task-specific AI agents by the end of this year, up from less than 5% today.
Translation: AI agents are about to start using your software the same way your staff do. The protocol that lets them do that is MCP.
What MCP actually is, in one paragraph
MCP is a standard interface. When a piece of software exposes an "MCP server," it's publishing a menu of things an AI agent can do with it. Read this report. Post this journal entry. Look up this client. Generate this form. The agent reads the menu, picks the right tool, and executes. Before MCP, every AI integration was a bespoke API project. After MCP, it's closer to plug-and-play.
The analogy I keep using with clients is USB. Devices used to ship with their own cables. Then USB happened. Vendors who didn't adopt it became hard to work with. Vendors who did adopt it got picked up by everything else.
Why this matters for accounting and SMSF firms
For a partner running a practice, the question is no longer whether you'll use AI. Some of your staff already are. The question is whether the software you spend tens of thousands of dollars on per year can be operated by an agent on behalf of a human, with an audit trail that holds up.
This connects to something we wrote about earlier this year: the Privacy Act AI disclosure requirement coming into force in December 2026. The firms that get ahead of that are the ones building controlled, documented AI workflows now. MCP is the infrastructure layer that makes those workflows auditable.
If your practice management system, document store, or SMSF admin platform doesn't expose an MCP server in 2026, a few things happen.
Your staff will start using agents that work around your software instead of with it. They'll copy data out, work in a side tool, and paste results back in. That's how shadow IT starts.
Your competitors using vendors with MCP support will move faster on the same work. Not dramatically faster on day one. Quietly faster, on the tasks that compound. Reconciliations. Document review. Fund admin checklists. Compliance forms.
And you'll find yourself locked into vendors whose roadmap doesn't match where the industry is heading. Switching costs will keep climbing.
What this looks like on the SMB side
For non-accounting SMB owners, the angle is different but related. If your CRM, your project tool, and your accounting package don't speak MCP, the AI agents you'll be running by mid-2027 (whether you planned to or not) won't be able to operate across them.
The good news is that the integration tax just dropped. Before MCP, connecting a new tool to an AI workflow meant a custom build for every pair. With MCP, it's one connection per tool. If you've ever paid for a Zapier or Make integration project that ran over budget, you'll feel this one in the budget line.
We've written about how AI automation fits into a small business in practice. The MCP shift changes the tooling layer underneath that picture, but the principles are the same: start with the workflow, then pick the tool.
The governance footnote
The honest version of this story includes a caveat. MCP is not finished. The 2026 roadmap from the maintainers explicitly lists audit trails, SSO-integrated auth, and gateway behaviour as the least defined of the four enterprise priorities. For tightly regulated environments, that matters.
So I'd push back on anyone telling accounting or SMSF firms to deploy agent-powered workflows on client data tomorrow. The protocol is real. The enterprise readiness work is still in progress. The right move for the next twelve months is to know what's coming, push your vendors on their roadmap, and pilot in low-risk areas.
The five questions to ask your software vendor
If you have a renewal coming up, or you're evaluating a new tool, these are the questions that separate vendors who get it from vendors who don't.
- Do you have an MCP server today? If not, what's the timeline?
- What operations are exposed through it? Read-only or read-write?
- How do you handle authentication, especially for agent-initiated actions?
- What audit trail is captured when an agent calls one of your tools?
- Are you tracking the standards work coming out of the Agentic AI Foundation under the Linux Foundation?
A vendor who can answer all five clearly is ahead of the curve. A vendor who hasn't thought about any of them is one you'll need to revisit in twelve months, and possibly replace.
What to do this quarter
If you're running a firm or a small business, three small actions go a long way.
Audit your current stack. List every piece of software your team uses daily. Mark which ones have public MCP support today (the vendor's docs or developer site will tell you). For the rest, send a single email to your account manager asking for the roadmap.
Pick one low-risk workflow. Something where an agent's mistakes are easy to catch. Document review. Draft reconciliations. First-pass compliance checks. Try it with a tool that already supports MCP and see what happens.
Decide on a review cadence. The protocol is evolving fast enough that an annual check is too slow. Quarterly is about right for the next year.
At OrionX we spend a lot of time looking at the gap between what vendors are shipping and what firms actually need. If you'd like a second opinion on your current stack and where the MCP-shaped holes are, get in touch. No deck, no pitch. Just an honest read.
Sources: Anthropic, original MCP announcement (November 2024) · Knak, MCP Adoption in 2026: What Marketers Need to Know · Truto, What is MCP: The 2026 Guide for SaaS PMs · David Soria Parra, The 2026 MCP Roadmap, modelcontextprotocol.io · Stacklok, State of Model Context Protocol in Software 2026 · CIO.com, Why Model Context Protocol is suddenly on every executive agenda
